Evaluating the Performance of Supervised Machine Learning Algorithms for Extracting Tree Canopy Cover

Thursday, September 19, 2024

9:30 AM – 10:30 AM MST

Location: McKee Hall

Poster Presenter(s)

Prajwol Babu Subedi, n/a

Graduate Research Assistant
University of Arkansas
Monticello, AR, Arkansas, United States

Submitter(s)

Hamdi Zurqani, n/a

Assistant Professor of Geospatial Science
University of Arkansas at Monticello - Monticello, AR
Monticello, Arkansas, United States

This study identifies Random Forest as the superior method in a comparison between six supervised Machine Learning (ML) algorithms. It provides enhanced accuracy through parameter optimization and diverse feature variable combinations, including the impact of algorithm choice on classification accuracy and the challenges of optimizing parameters for different environmental contexts.

Learning Objectives:

Upon completion, participants will be able to thoroughly identify and quantify the impacts of parameter optimization on the accuracy of supervised machine learning (ML) algorithms used in remote sensing-based image classifications. They will understand how varying the sizes of training datasets influence model robustness and generalization capabilities. Additionally, they will explore the effects of different feature variable combinations, including spectral bands and textural indices, on algorithmic precision. Participants will critically evaluate these factors to understand their individual and combined influence on the performance of algorithms like Random Forest (RF) within the context of image classification and environmental monitoring tasks.